My work focuses on how we can identify and address undiagnosed heterogeneity in samples of heterogeneous time-series by drawing on techniques from graph theory and network analysis.

I also have a line of work directly in network analysis using cascading failure models and fuzzy clustering methods.

whether a failure of a specific vertex results in overwhelming failure throughout a graph or whether the pattern of failure is similar to any other randomly selected vertex.

Adding to (1.), we conduct simulations where peripheral vertices are more influential to one another and ones where hubs are highly connected *and* influential. In the former, we can recreate the results from this paper and in the latter we find that the topology of the graph is a large player in

2. We don't allow flipped vertices to come back into the system; so, failure in our simulations is permanent.

This could be a time-scale difference in the mode of collapse. Some systems fail so quickly that other variables cannot react while in slower time-scales, this is more plausible

Thanks for the shout out, @omidvebrahimi.bsky.social. This is an interesting paper!

Our simulations are a bit different. So, the type of world our models describe are a bit different and I can describe below:

1. We assumed that the connections between vertices are weighted and heterogeneous.

Thank you, Cam! 🥹
Helps to be a part of a stellar department too 😉

“Among several undergraduates I've worked with over the years, he is definitely in the top 20 (N = 20)."

Thanks, Siwei!

I wouldn't have been able to complete this work without mentors and collaborators:
- Sy-Miin Chow
- Peter Molenaar
- @fishingwithzack.bsky.social
- Michael Hunter
- @chadshenkphd.bsky.social
- Michael Russell

In the paper, we highlight the strengths of modeling in continuous-time and contrast it with modeling in discrete-time dynamic networks.

We also highlight some key weaknesses in implementing an automated search of continuous-time dynamic networks RE: initial conditions and determining them sensibly

I think you’re in charge of adding people if you made the thread! I am also new and don’t know anything hahah

Thanks, Björn!
Looks like I’m too late to jump on that starter pack haha!

Hello! I’m primarily doing work in dynamic network modeling and community detection. Could I be added to this? Thanks for putting this together!

Thanks, Björn! Glad you liked it; hope the reviewers do too haha

We really wanted to be clear during the empirical application that we didn't magically solve issues with starting values in continuous-time. These systems are just so much more sensitive than discrete-time ones.

Special thanks to my dissertation committee members:
@fishingwithzack.bsky.social, Mike Hunter, Chad Shenk, Mike Russell, and my advisors Sy-Miin Chow and Peter Molenaar for their help and guidance throughout my PhD.

I could not have done this without all of you!

We also tested ct-gimme on real-world data and comment on some issues that researchers fitting continuous-time models are still likely to face even with the automated behavior of ct-gimme.

We evaluated the performance of what we're calling ct-gimme in simulations and found that it outperforms N = 1 fitting in continuous-time by leveraging information across the sample prior to individual model fitting as per traditional GIMME